Young, Clark, Goffus, and Hoane (Learning and Motivation, 40(2), 160–177, 2009) documented significant advantages of linear and nonlinear mixed-effects modeling in the analysis of Morris water maze data. However, they also noted a caution regarding the impact of the common practice of ending a trial when the rat had not reached the platform by a preestablished deadline. The present study revisits their conclusions by considering a new approach that involves multilevel (i.e., mixed effects) censored generalized linear regression using Bayesian analysis. A censored regression explicitly models the censoring created by prematurely ending a trial, and the use of generalized linear regression incorporates the skewed distribution of latency data as well as the nonlinear relationships this can produce. This approach is contrasted with a standard multilevel linear and nonlinear regression using two case studies. The censored generalized linear regression better models the observed relationships, but the linear regression created mixed results and clearly resulted in model misspecification.

Young、Clark、Goffus 和 Hoane（学习和动机，40(2), 160–177, 2009) 记录了线性和非线性混合效应建模在莫里斯水迷宫数据分析中的显着优势。然而，他们还注意到，当大鼠未在预定的截止日期前到达平台时，结束试验的常见做法的影响需要谨慎。本研究通过考虑涉及使用贝叶斯分析的多级（即混合效应）删失广义线性回归的新方法来重新审视他们的结论。删失回归明确模拟过早结束试验造成的删失，并且广义线性回归的使用结合了延迟数据的偏态分布以及由此可能产生的非线性关系。这种方法与使用两个案例研究的标准多级线性和非线性回归进行对比。